@HarryStebbings@toddsaunders Don't have enough money to make it interesting for you but I would bet against you with the CDTM.
More unicorn valuations within the next 10 years with a founder from the CDTM than from Project Europe?
@paulg@Heliocene in practice, you sign an additional document that delegates your presence to the presence of the attorney of either the lead investor or the startup.
The readings happen but almost no one is there.
I hope this makes you reconsider to still invest into german startups ;)
@nejatian@Opendoor A question: When you are trying to think through a difficult problem, are you defaulting to AI first for breaking it down and thinking through it? Or first think it through once yourself and then go to AI for feedback/improvements?
@mckaywrigley why exactly do you like claude opus better?
I feel like for nuanced issues & debugging it was more accurate and less "all over the place". Haven't used it for bigger new features though.
It did feel even slower than Opus in those cases, probably because it had to think a lot.
@shl@jackfriks I think they lost quite a bit of trust by drastically overselling early on and then going viral with it... hard to get rid of that first impression for a lot of people
@jasonlk@Replit It is really cool to follow along and impressive how all in you are. Also love the video game analogy from 1-2 days ago. I feel like tilt as a gaming concept is also very real for vibe coding
@RichardSocher@karpathy Could be that I selected the wrong thing in the funnel. I definitely selected a "Research Agent" but maybe that was the wrong one
@RichardSocher@karpathy I've tried based on this post but it didn't work for me at all.
Had to re-start 3 times with the follow-up questions due to bad UX. The first draft was WAY to shallow and then when I asked for a more fleshed out draft it always stopped halfway due to message length.
@Shpigford@maybe@rails Can I somewhere read about the specific issues around next and SOC2 and SEC compliance? Like a specific GitHub issue or something?
@heikirk I’d guess even more than half… if you are mainly working with text and want to preselect context from a large body of text the classic index part is just more important than the vector